Comparing Schemas for Spatiotemporal Data Warehouses
Author
Domínguez, Luis
Term
4. term
Education
Publication year
2003
Abstract
At spore biler for at analysere trafik er en lokationsbaseret tjeneste, der skaber meget store datamængder. Denne datavolumen er typisk for stor til, at klassiske OLTP-systemer—databaser optimeret til mange små transaktioner—kan håndtere den effektivt til analyse. Denne afhandling opbygger et spatio-temporalt datalager (et lager, der organiserer data efter både sted og tid) for bevægelige objekter og vurderer flere alternative design. Datalageret gemmer fakta som bilers positioner (X- og Y-koordinater knyttet til en vejstrækning) med tidsstempler, vejudnyttelse samt længde og varighed af ture. Det fyldes fra to kilder: virkelige GPS-data fra kørende biler og syntetiske data fra en generator. For at forbedre svartider oprettes indeks og materialiserede visninger (forhåndsberegnede sammenfatninger). Et sæt typiske forespørgsler afvikles derefter for at måle ydeevne og sammenligne designene.
Tracking vehicles to analyze traffic is a location-based service that generates very large amounts of data. This volume is typically too big for classical OLTP systems—databases optimized for many small transactions—to handle efficiently for analysis. This thesis builds a spatiotemporal data warehouse (a data store organized by both location and time) for moving objects and evaluates several alternative designs. The warehouse stores facts such as car positions (X and Y coordinates linked to a road segment) with timestamps, road utilization, and the length and duration of trips. It is populated from two sources: real GPS data from moving cars and synthetic data from a generator. To improve query speed, the warehouse is enhanced with indexes and materialized views (precomputed summaries). A set of typical traffic queries is then executed to measure performance and compare the designs.
[This abstract was generated with the help of AI]
Documents